Computerized system and method for predicting morality risk using a Lyapunov stability classifier

Abstract

A method and system suitable for automated surveillance of intensive care unit patients for information denoting likelihood of in-hospital survival or mortality, represented in the timeseries of scoring systems such as APACHE III. Techniques from digital signal processing and Lyapunov stability analysis are combined in a method that allows for optimization of statistical hypothesis testing that is robust against short time series of as few as five time points. Once optimized, the method and system can achieve high-sensitivity high-specificity classification of survivorship, while avoiding false-positive prediction of mortality.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS[0001]This application claims the benefit of priority to U.S. application Ser. No. 10 / 751,821 filed on Jan. 5, 2004 which in turn claims the benefit of priority to U.S. Provisional Application No. 60 / 468,765 filed May 8, 2003.STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT[0002]Not applicable.TECHNICAL FIELD[0003]The present invention relates to a system and method for identifying which individual persons admitted to an intensive care unit (ICU) are likely to die prior to hospital discharge, and which persons are likely to recover and survive to discharge.BACKGROUND OF THE INVENTION[0004]A very difficult clinical problem facing clinicians is knowing when further treatment is futile and no longer appropriate in a patient who has developed severe complications after surgery and is being treated in an ICU. It is now possible to prolong the process of dying among such patients. This results in unnecessary pain and loss of dignity fo...

AI Technical Summary

Benefits of technology

The present invention provides a method and system for predicting mortality risk using a Lyapunov stability classifier. This method involves accessing mortality-predictive serial data, performing spectral analysis, and calculating the Lyapunov exponent. If the Lyapunov exponent is negative, outputting values for the exponent for a plurality of times in the timeseries. This invention can help improve the accuracy of predicting a person's risk of dying in the hospital.

Problems solved by technology

A very difficult clinical problem facing clinicians is knowing when further treatment is futile and no longer appropriate in a patient who has developed severe complications after surgery and is being treated in an ICU.

This results in unnecessary pain and loss of dignity for the patient, anguish and distress for the patient's relatives and is dehumanizing for the clinical and nursing staff.

The complementary clinical problem facing clinicians, that of knowing when further aggressive treatment is appropriate and has reasonable odds of saving the patient's life and yielding quality of life and other benefits in addition to survival, is equally difficult.

Prognostic criteria based on “static” analysis of group statistics are of little value in decisions to withhold or withdraw therapy from ICU patients too ill to benefit, since they do not provide adequate information on the features that distinguish non-survivors from survivors.

The feeling that a patient admitted to the ICU should not be considered as a ‘terminal patient’ produces difficulties associated with rationalizing the use of new and costly technologies in a situation where resources are scarce, such as the modern healthcare systems.

Over the last 15 years, several systems for the assessment of multiorgan failure have been developed, mainly because multiorgan failure represents the main cause of mortality and morbidity among critically ill patients managed in the ICU.

More recent trend analysis techniques such as SOFA (‘sequential organ failure assessment’, see Cook 2001, Hutchinson 2000, Pettila 2002, Rosenberg 2002) continue to suffer from inadequate sensitivity.

Other algorithms using serial scoring have had unacceptably high false-positive rates.

If used prospectively, this algorithm does have the potential to indicate the futility of continued intensive care but at the high cost of nearly 1 in 20 patients who would survive if intensive care were continued.

Chang's and others' approaches were unsatisfactory in terms of excessive reliance on long timeseries (74% of predictions resolvable by trend analysis required data from seven days or more in the ICU, too long to be of significant help in the contemporary situation with its focus on prompt decision-making soon after admission and aggressive discharge-planning).

In part, the failure of the prior art can be traced to inappropriate pooling of data from groups of patients with markedly differing mortality rates, including many whose probability of in-hospital death was low or moderate.

Images